| The proposal of "Industry 4.0" and "Made in China 2025" makes intelligent manufacturing the main direction of future industrial development.Intelligent shop scheduling is an important research content of intelligent manufacturing.Existing studies on shop scheduling mainly focus on meta-heuristic algorithms.However,the results of meta-heuristic algorithms are unstable.Besides,the performance of meta-heuristic algorithms heavily depends on the specific problem and human experience.Therefore,the research and implementation of the reinforcement learning based shop scheduling system are of great research significance and application value.The main work of this paper is as follows.1.This paper mainly studies the flow shop scheduling problem with batch processing machines(BPM-FSP).The BPM-FSP combines the job batching problem and the job scheduling problem,which is more complex and more oriented to actual production.Firstly,this paper establishes a mathematical model of the BPM-FSP problem to minimize the maximum completion time.This paper formulates the BPM-FSP scheduling process as a Markov Decision Process(MDP)and designs the state,action,and reward respectively.Besides,the BPM-FSP problem is regarded as a special sequence-to-sequence problem and a basic scheduling framework based on the pointer network is proposed.Additionally,this paper establishes a BPM-FSP solution model,which consists of a job batching module and a job scheduling module.2.Based on the proposed basic scheduling framework,this paper designs the job batching model and the job scheduling model for the job batching module and the job scheduling module respectively.Further,this paper proposes the job batching algorithm and the job scheduling algorithm for model training based on the deep reinforcement learning method.Finally,this paper analyzes the convergence of the proposed models under different parameter settings and compares the proposed scheduling method based on deep reinforcement learning with other methods on different sizes of the public data set and the actual production data set.The experimental results indicate that the proposed shop scheduling method can better solve the BPM-FSP problem,especially in large-scale scheduling scenarios.3.This paper designs and implements the shop scheduling system based on deep reinforcement learning.As a significant component of the Industrial Internet Intelligent Cloud Collaboration Platform,this system has been successfully demonstrated and applied in the workshop of the wide and thick plate coil Plant of Nanjing Iron and Steel Co.,LTD in Jiangsu Province to effectively assist the enterprise in production scheduling. |
